Research On Pose-oblivious Retrieval Of 3D Animation Material Model

We present an approach for robust shape retrieval form databases containing articulated 3D models which is used largely in the computer animation field. First, each shape is represented by the eigenvectors of an appropriately defined affinity matrix based on approximated geodesic distance, forming a spectral embedding which achieves normalization against rigid-body transformations, uniform scaling, and shape articulation (i.e., pose changing). This procedure is called the pose-oblivious transform. We need to reconstruct an adjacent graph for the non-connectivity models. Second, retrieval is performed in the spectral domain using global shape descriptors, including eigenvalue-based descriptor, histogram-based descriptor, key-point correspondence based descriptor and the hierarchical descriptor consisting of the above three. Last, we implement an prototype system and integrated it with another effective and creative animation design system. On the database of articulated 3D shapes, the descriptors based on spectral approach leads to an absolute improvement in retrieval performance. The best retrieval results are obtained using the hierarchical descriptor. Also, it's very efficient and fit for on-line 3D shape retrieval.